Face Generation

In this project, you'll define and train a DCGAN on a dataset of faces. Your goal is to get a generator network to generate new images of faces that look as realistic as possible!

The project will be broken down into a series of tasks from loading in data to defining and training adversarial networks. At the end of the notebook, you'll be able to visualize the results of your trained Generator to see how it performs; your generated samples should look like fairly realistic faces with small amounts of noise.

Get the Data

You'll be using the CelebFaces Attributes Dataset (CelebA) to train your adversarial networks.

This dataset is more complex than the number datasets (like MNIST or SVHN) you've been working with, and so, you should prepare to define deeper networks and train them for a longer time to get good results. It is suggested that you utilize a GPU for training.

Pre-processed Data

Since the project's main focus is on building the GANs, we've done some of the pre-processing for you. Each of the CelebA images has been cropped to remove parts of the image that don't include a face, then resized down to 64x64x3 NumPy images. Some sample data is show below.

If you are working locally, you can download this data by clicking here

This is a zip file that you'll need to extract in the home directory of this notebook for further loading and processing. After extracting the data, you should be left with a directory of data processed_celeba_small/

In [1]:
# can comment out after executing
# !unzip processed_celeba_small.zip
In [2]:
data_dir = 'processed_celeba_small/'

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import pickle as pkl
import matplotlib.pyplot as plt
import numpy as np
import problem_unittests as tests
#import helper

%matplotlib inline

Visualize the CelebA Data

The CelebA dataset contains over 200,000 celebrity images with annotations. Since you're going to be generating faces, you won't need the annotations, you'll only need the images. Note that these are color images with 3 color channels (RGB)#RGB_Images) each.

Pre-process and Load the Data

Since the project's main focus is on building the GANs, we've done some of the pre-processing for you. Each of the CelebA images has been cropped to remove parts of the image that don't include a face, then resized down to 64x64x3 NumPy images. This pre-processed dataset is a smaller subset of the very large CelebA data.

There are a few other steps that you'll need to transform this data and create a DataLoader.

Exercise: Complete the following get_dataloader function, such that it satisfies these requirements:

  • Your images should be square, Tensor images of size image_size x image_size in the x and y dimension.
  • Your function should return a DataLoader that shuffles and batches these Tensor images.

ImageFolder

To create a dataset given a directory of images, it's recommended that you use PyTorch's ImageFolder wrapper, with a root directory processed_celeba_small/ and data transformation passed in.

In [3]:
# necessary imports
import torch
from torchvision import datasets
from torchvision import transforms
In [4]:
def get_dataloader(batch_size, image_size, data_dir='processed_celeba_small/'):
    """
    Batch the neural network data using DataLoader
    :param batch_size: The size of each batch; the number of images in a batch
    :param img_size: The square size of the image data (x, y)
    :param data_dir: Directory where image data is located
    :return: DataLoader with batched data
    """
    
    transform = transforms.Compose([
        transforms.Resize(image_size),
        transforms.CenterCrop(image_size),
        transforms.ToTensor()
    ])
    
    dataset = datasets.ImageFolder(data_dir, transform=transform)
    
    # TODO: Implement function and return a dataloader
    data_loader = torch.utils.data.DataLoader(dataset=dataset,
                                             batch_size=batch_size,
                                             shuffle=True)
    
    return data_loader

Create a DataLoader

Exercise: Create a DataLoader celeba_train_loader with appropriate hyperparameters.

Call the above function and create a dataloader to view images.

  • You can decide on any reasonable batch_size parameter
  • Your image_size must be 32. Resizing the data to a smaller size will make for faster training, while still creating convincing images of faces!
In [5]:
# Define function hyperparameters
batch_size = 128
img_size = 32

"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
# Call your function and get a dataloader
celeba_train_loader = get_dataloader(batch_size, img_size)

Next, you can view some images! You should seen square images of somewhat-centered faces.

Note: You'll need to convert the Tensor images into a NumPy type and transpose the dimensions to correctly display an image, suggested imshow code is below, but it may not be perfect.

In [6]:
# helper display function
def imshow(img):
    npimg = img.numpy()
    plt.imshow(np.transpose(npimg, (1, 2, 0)))

"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
# obtain one batch of training images
dataiter = iter(celeba_train_loader)
images, _ = dataiter.next() # _ for no labels

# plot the images in the batch, along with the corresponding labels
fig = plt.figure(figsize=(20, 4))
plot_size=20
for idx in np.arange(plot_size):
    ax = fig.add_subplot(2, plot_size/2, idx+1, xticks=[], yticks=[])
    imshow(images[idx])

Exercise: Pre-process your image data and scale it to a pixel range of -1 to 1

You need to do a bit of pre-processing; you know that the output of a tanh activated generator will contain pixel values in a range from -1 to 1, and so, we need to rescale our training images to a range of -1 to 1. (Right now, they are in a range from 0-1.)

In [7]:
# TODO: Complete the scale function
def scale(x, feature_range=(-1, 1)):
    ''' Scale takes in an image x and returns that image, scaled
       with a feature_range of pixel values from -1 to 1. 
       This function assumes that the input x is already scaled from 0-1.'''
    # assume x is scaled to (0, 1)
    # scale to feature_range and return scaled x
    min, max = feature_range
    x = x * (max - min) + min
    return x
In [8]:
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
# check scaled range
# should be close to -1 to 1
img = images[0]
scaled_img = scale(img)

print('Min: ', scaled_img.min())
print('Max: ', scaled_img.max())
Min:  tensor(-1.)
Max:  tensor(0.7412)

Define the Model

A GAN is comprised of two adversarial networks, a discriminator and a generator.

Discriminator

Your first task will be to define the discriminator. This is a convolutional classifier like you've built before, only without any maxpooling layers. To deal with this complex data, it's suggested you use a deep network with normalization. You are also allowed to create any helper functions that may be useful.

Exercise: Complete the Discriminator class

  • The inputs to the discriminator are 32x32x3 tensor images
  • The output should be a single value that will indicate whether a given image is real or fake
In [9]:
import torch.nn as nn
import torch.nn.functional as F
In [10]:
from collections import OrderedDict
In [11]:
class Discriminator(nn.Module):

    def __init__(self, conv_dim):
        """
        Initialize the Discriminator Module
        :param conv_dim: The depth of the first convolutional layer
        """
        super(Discriminator, self).__init__()

        self.conv_dim = conv_dim
        
        kernel_size = 4 
        stride = 2
        padding = 1
        
        self.features = nn.Sequential(OrderedDict([
            ('conv1', nn.Conv2d(3, conv_dim,kernel_size,stride,padding,bias=False)),
            ('lrelu1', nn.LeakyReLU(negative_slope=0.2,inplace=True)),
            ('conv2', nn.Conv2d(conv_dim, conv_dim*2,kernel_size,stride,padding,bias=False)),
            ('bn2', nn.BatchNorm2d(conv_dim*2)),
            ('lrelu2', nn.LeakyReLU(negative_slope=0.2,inplace=True)),
            ('conv3', nn.Conv2d(conv_dim*2, conv_dim*4,kernel_size,stride,padding,bias=False)),
            ('bn3', nn.BatchNorm2d(conv_dim*4)),
            ('lrelu3', nn.LeakyReLU(negative_slope=0.2,inplace=True))
        ]))
        
        self.classifier = nn.Linear(conv_dim*4*4*4,1)
        

    def forward(self, x):
        """
        Forward propagation of the neural network
        :param x: The input to the neural network     
        :return: Discriminator logits; the output of the neural network
        """
        out = self.features(x)
        # flatten
        out = out.view(-1, self.conv_dim*4*4*4)
        # final output layer
        out = self.classifier(out)
        return out

    
print(Discriminator(32))
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_discriminator(Discriminator)
Discriminator(
  (features): Sequential(
    (conv1): Conv2d(3, 32, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1), bias=False)
    (lrelu1): LeakyReLU(negative_slope=0.2, inplace)
    (conv2): Conv2d(32, 64, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1), bias=False)
    (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (lrelu2): LeakyReLU(negative_slope=0.2, inplace)
    (conv3): Conv2d(64, 128, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1), bias=False)
    (bn3): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (lrelu3): LeakyReLU(negative_slope=0.2, inplace)
  )
  (classifier): Linear(in_features=2048, out_features=1, bias=True)
)
Tests Passed

Generator

The generator should upsample an input and generate a new image of the same size as our training data 32x32x3. This should be mostly transpose convolutional layers with normalization applied to the outputs.

Exercise: Complete the Generator class

  • The inputs to the generator are vectors of some length z_size
  • The output should be a image of shape 32x32x3
In [12]:
class Generator(nn.Module):
    
    def __init__(self, z_size, conv_dim):
        """
        Initialize the Generator Module
        :param z_size: The length of the input latent vector, z
        :param conv_dim: The depth of the inputs to the *last* transpose convolutional layer
        """
        super(Generator, self).__init__()
        
        self.conv_dim = conv_dim
        
        # first, fully-connected layer
        self.fc = nn.Linear(z_size, conv_dim*4*4*4)
        
        kernel_size = 4
        stride = 2
        padding = 1
        
        self.generator = nn.Sequential(OrderedDict([
            ('t_conv1', nn.ConvTranspose2d(conv_dim*4, conv_dim*2, kernel_size,
                                          stride, padding, bias=False)),
            ('bn1', nn.BatchNorm2d(conv_dim*2)),
            ('relu1', nn.ReLU(True)),
            ('t_conv2', nn.ConvTranspose2d(conv_dim*2, conv_dim, kernel_size,
                                          stride, padding, bias=False)),
            ('bn2', nn.BatchNorm2d(conv_dim)),
            ('relu2', nn.ReLU(True)),
            ('t_conv3', nn.ConvTranspose2d(conv_dim, 3, kernel_size,
                                          stride, padding, bias=False)),
            ('tanh3', nn.Tanh())
        ]))
        

    def forward(self, x):
        """
        Forward propagation of the neural network
        :param x: The input to the neural network     
        :return: A 32x32x3 Tensor image as output
        """
        
        out = self.fc(x)
        
        # reshape
        out = out.view(-1, self.conv_dim*4, 4, 4)# (batch_size, depth, 4, 4)
        
        out = self.generator(out)
        
        return out

print(Generator(100,32))
    
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_generator(Generator)
Generator(
  (fc): Linear(in_features=100, out_features=2048, bias=True)
  (generator): Sequential(
    (t_conv1): ConvTranspose2d(128, 64, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1), bias=False)
    (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (relu1): ReLU(inplace)
    (t_conv2): ConvTranspose2d(64, 32, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1), bias=False)
    (bn2): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (relu2): ReLU(inplace)
    (t_conv3): ConvTranspose2d(32, 3, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1), bias=False)
    (tanh3): Tanh()
  )
)
Tests Passed

Initialize the weights of your networks

To help your models converge, you should initialize the weights of the convolutional and linear layers in your model. From reading the original DCGAN paper, they say:

All weights were initialized from a zero-centered Normal distribution with standard deviation 0.02.

So, your next task will be to define a weight initialization function that does just this!

You can refer back to the lesson on weight initialization or even consult existing model code, such as that from the networks.py file in CycleGAN Github repository to help you complete this function.

Exercise: Complete the weight initialization function

  • This should initialize only convolutional and linear layers
  • Initialize the weights to a normal distribution, centered around 0, with a standard deviation of 0.02.
  • The bias terms, if they exist, may be left alone or set to 0.
In [13]:
from torch.nn import init

def weights_init_normal(m):
    """
    Applies initial weights to certain layers in a model .
    The weights are taken from a normal distribution 
    with mean = 0, std dev = 0.02.
    :param m: A module or layer in a network    
    """
    # classname will be something like:
    # `Conv`, `BatchNorm2d`, `Linear`, etc.
    classname = m.__class__.__name__
    if hasattr(m, 'weight') and (classname.find('Conv') != -1 or classname.find('Linear') != -1):
        init.normal_(m.weight.data, 0.0, 0.02)
        
        if hasattr(m, 'bias') and m.bias is not None:
                init.constant_(m.bias.data, 0.0)
                
    elif classname.find('BatchNorm2d') != -1:
        init.normal_(m.weight.data, 1.0, 0.02)
        init.constant_(m.bias.data, 0.0)  

Build complete network

Define your models' hyperparameters and instantiate the discriminator and generator from the classes defined above. Make sure you've passed in the correct input arguments.

In [14]:
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
def build_network(d_conv_dim, g_conv_dim, z_size):
    # define discriminator and generator
    D = Discriminator(d_conv_dim)
    G = Generator(z_size=z_size, conv_dim=g_conv_dim)

    # initialize model weights
    D.apply(weights_init_normal)
    G.apply(weights_init_normal)

    print(D)
    print()
    print(G)
    
    return D, G

Exercise: Define model hyperparameters

In [15]:
# Define model hyperparams
d_conv_dim = 32
g_conv_dim = 32
z_size = 100

"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
D, G = build_network(d_conv_dim, g_conv_dim, z_size)
Discriminator(
  (features): Sequential(
    (conv1): Conv2d(3, 32, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1), bias=False)
    (lrelu1): LeakyReLU(negative_slope=0.2, inplace)
    (conv2): Conv2d(32, 64, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1), bias=False)
    (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (lrelu2): LeakyReLU(negative_slope=0.2, inplace)
    (conv3): Conv2d(64, 128, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1), bias=False)
    (bn3): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (lrelu3): LeakyReLU(negative_slope=0.2, inplace)
  )
  (classifier): Linear(in_features=2048, out_features=1, bias=True)
)

Generator(
  (fc): Linear(in_features=100, out_features=2048, bias=True)
  (generator): Sequential(
    (t_conv1): ConvTranspose2d(128, 64, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1), bias=False)
    (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (relu1): ReLU(inplace)
    (t_conv2): ConvTranspose2d(64, 32, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1), bias=False)
    (bn2): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (relu2): ReLU(inplace)
    (t_conv3): ConvTranspose2d(32, 3, kernel_size=(4, 4), stride=(2, 2), padding=(1, 1), bias=False)
    (tanh3): Tanh()
  )
)

Training on GPU

Check if you can train on GPU. Here, we'll set this as a boolean variable train_on_gpu. Later, you'll be responsible for making sure that

  • Models,
  • Model inputs, and
  • Loss function arguments

Are moved to GPU, where appropriate.

In [16]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import torch

# Check for a GPU
train_on_gpu = torch.cuda.is_available()
if not train_on_gpu:
    print('No GPU found. Please use a GPU to train your neural network.')
else:
    print('Training on GPU!')
Training on GPU!

Discriminator and Generator Losses

Now we need to calculate the losses for both types of adversarial networks.

Discriminator Losses

  • For the discriminator, the total loss is the sum of the losses for real and fake images, d_loss = d_real_loss + d_fake_loss.
  • Remember that we want the discriminator to output 1 for real images and 0 for fake images, so we need to set up the losses to reflect that.

Generator Loss

The generator loss will look similar only with flipped labels. The generator's goal is to get the discriminator to think its generated images are real.

Exercise: Complete real and fake loss functions

You may choose to use either cross entropy or a least squares error loss to complete the following real_loss and fake_loss functions.

In [17]:
def real_loss(D_out, smooth = False):
    '''Calculates how close discriminator outputs are to being real.
       param, D_out: discriminator logits
       return: real loss'''
    batch_size = D_out.size(0)
    # label smoothing
    if smooth:
        # smooth, real labels = 0.9
        labels = torch.ones(batch_size)*0.9
    else:
        labels = torch.ones(batch_size) # real labels = 1
    # move labels to GPU if available     
    if train_on_gpu:
        labels = labels.cuda()
    # binary cross entropy with logits loss
    criterion = nn.BCEWithLogitsLoss()
    # calculate loss
    loss = criterion(D_out.squeeze(), labels)
    return loss


def fake_loss(D_out):
    '''Calculates how close discriminator outputs are to being fake.
       param, D_out: discriminator logits
       return: fake loss'''
    batch_size = D_out.size(0)
    labels = torch.zeros(batch_size) # fake labels = 0
    if train_on_gpu:
        labels = labels.cuda()
    criterion = nn.BCEWithLogitsLoss()
    # calculate loss
    loss = criterion(D_out.squeeze(), labels)
    return loss 

Optimizers

Exercise: Define optimizers for your Discriminator (D) and Generator (G)

Define optimizers for your models with appropriate hyperparameters.

In [18]:
import torch.optim as optim

# params
lr = 0.0002
beta1=0.5
beta2=0.999 # default value

# Create optimizers for the discriminator and generator
d_optimizer = optim.Adam(D.parameters(), lr, [beta1, beta2])
g_optimizer = optim.Adam(G.parameters(), lr, [beta1, beta2])

Training

Training will involve alternating between training the discriminator and the generator. You'll use your functions real_loss and fake_loss to help you calculate the discriminator losses.

  • You should train the discriminator by alternating on real and fake images
  • Then the generator, which tries to trick the discriminator and should have an opposing loss function

Saving Samples

You've been given some code to print out some loss statistics and save some generated "fake" samples.

Exercise: Complete the training function

Keep in mind that, if you've moved your models to GPU, you'll also have to move any model inputs to GPU.

In [19]:
import time
def train(D, G, n_epochs, print_every=500):
    '''Trains adversarial networks for some number of epochs
       param, D: the discriminator network
       param, G: the generator network
       param, n_epochs: number of epochs to train for
       param, print_every: when to print and record the models' losses
       return: D and G losses
    '''
    
    # move models to GPU
    if train_on_gpu:
        D.cuda()
        G.cuda()

    # keep track of loss and generated, "fake" samples
    samples = []
    losses = []

    # Get some fixed data for sampling. These are images that are held
    # constant throughout training, and allow us to inspect the model's performance
    sample_size=16
    fixed_z = np.random.uniform(-1, 1, size=(sample_size, z_size))
    fixed_z = torch.from_numpy(fixed_z).float()
    # move z to GPU if available
    if train_on_gpu:
        fixed_z = fixed_z.cuda()
    
    # epoch training loop
    for epoch in range(n_epochs):
        t1 = time.time()
        # batch training loop
        for batch_i, (real_images, _) in enumerate(celeba_train_loader):

            batch_size = real_images.size(0)
            real_images = scale(real_images)

            # ===============================================
            #         YOUR CODE HERE: TRAIN THE NETWORKS
            # ===============================================
            
            d_optimizer.zero_grad()          
            
            # 1. Train the discriminator on real and fake images
            
            # Compute the discriminator losses on real images 
            if train_on_gpu:
                real_images = real_images.cuda()
                
            D_real = D(real_images)
            d_real_loss = real_loss(D_real)
            
            # Generate fake images
            z = np.random.uniform(-1, 1, size=(batch_size, z_size))
            z = torch.from_numpy(z).float()
            # move x to GPU, if available
            if train_on_gpu:
                z = z.cuda()
            fake_images = G(z)
            
            # Compute the discriminator losses on fake images            
            D_fake = D(fake_images)
            d_fake_loss = fake_loss(D_fake)
            
            # add up loss and perform backprop
            d_loss = d_real_loss + d_fake_loss
            d_loss.backward()
            d_optimizer.step()

            # 2. Train the generator with an adversarial loss
            
            g_optimizer.zero_grad()
            
            # Generate fake images
            z = np.random.uniform(-1, 1, size=(batch_size, z_size))
            z = torch.from_numpy(z).float()
            if train_on_gpu:
                z = z.cuda()
            fake_images = G(z)
            
            # Compute the discriminator losses on fake images 
            # using flipped labels!
            D_fake = D(fake_images)
            g_loss = real_loss(D_fake) # use real loss to flip labels
            
            # perform backprop
            g_loss.backward()
            g_optimizer.step()
            
            # ===============================================
            #              END OF YOUR CODE
            # ===============================================

            # Print some loss stats
            if batch_i % print_every == 0:
                # append discriminator loss and generator loss
                losses.append((d_loss.item(), g_loss.item()))
                # print discriminator and generator loss
                
                print('Epoch [{:5d}/{:5d}] Time: {:2.2f}min | d_loss: {:6.4f} | g_loss: {:6.4f}'.format(
                        epoch+1, n_epochs, (time.time() - t1) / 60, d_loss.item(), g_loss.item()))


        ## AFTER EACH EPOCH##    
        # this code assumes your generator is named G, feel free to change the name
        # generate and save sample, fake images
        G.eval() # for generating samples
        samples_z = G(fixed_z)
        samples.append(samples_z)
        
        # view samples(defined below... will give error now.. just using to see how model showing images)
        _ = view_samples(-1, samples)
        
        G.train() # back to training mode

    # Save training generator samples
    with open('train_samples.pkl', 'wb') as f:
        pkl.dump(samples, f)
    
    # finally return losses
    return losses

Set your number of training epochs and train your GAN!

In [21]:
# set number of epochs 
n_epochs = 50


"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
# call training function
losses = train(D, G, n_epochs=n_epochs)
Epoch [    1/   50] Time: 0.00min | d_loss: 1.8281 | g_loss: 0.5106
Epoch [    1/   50] Time: 0.90min | d_loss: 0.9815 | g_loss: 1.1560
Epoch [    2/   50] Time: 0.00min | d_loss: 0.8845 | g_loss: 1.3274
Epoch [    2/   50] Time: 0.89min | d_loss: 1.1839 | g_loss: 1.3852
Epoch [    3/   50] Time: 0.00min | d_loss: 1.1640 | g_loss: 1.4949
Epoch [    3/   50] Time: 0.89min | d_loss: 1.0753 | g_loss: 1.2674
Epoch [    4/   50] Time: 0.00min | d_loss: 1.1332 | g_loss: 1.7640
Epoch [    4/   50] Time: 0.89min | d_loss: 1.1140 | g_loss: 1.6489
Epoch [    5/   50] Time: 0.00min | d_loss: 1.8538 | g_loss: 2.2552
Epoch [    5/   50] Time: 0.89min | d_loss: 1.1226 | g_loss: 0.8063
Epoch [    6/   50] Time: 0.00min | d_loss: 1.0813 | g_loss: 1.0534
Epoch [    6/   50] Time: 0.89min | d_loss: 0.8473 | g_loss: 1.3951
Epoch [    7/   50] Time: 0.00min | d_loss: 0.8688 | g_loss: 1.2566
Epoch [    7/   50] Time: 0.89min | d_loss: 0.9317 | g_loss: 1.1284
Epoch [    8/   50] Time: 0.00min | d_loss: 0.7776 | g_loss: 1.3923
Epoch [    8/   50] Time: 0.89min | d_loss: 0.8595 | g_loss: 0.9607
Epoch [    9/   50] Time: 0.00min | d_loss: 0.8340 | g_loss: 1.6079
Epoch [    9/   50] Time: 0.89min | d_loss: 0.8148 | g_loss: 1.1706
Epoch [   10/   50] Time: 0.00min | d_loss: 1.1527 | g_loss: 3.0040
Epoch [   10/   50] Time: 0.89min | d_loss: 0.7914 | g_loss: 1.9976
Epoch [   11/   50] Time: 0.00min | d_loss: 0.6794 | g_loss: 1.9704
Epoch [   11/   50] Time: 0.88min | d_loss: 1.0077 | g_loss: 0.8372
Epoch [   12/   50] Time: 0.00min | d_loss: 0.9264 | g_loss: 2.1708
Epoch [   12/   50] Time: 0.89min | d_loss: 0.7560 | g_loss: 1.3663
Epoch [   13/   50] Time: 0.00min | d_loss: 0.9804 | g_loss: 2.3708
Epoch [   13/   50] Time: 0.89min | d_loss: 0.7173 | g_loss: 1.8956
Epoch [   14/   50] Time: 0.00min | d_loss: 0.6743 | g_loss: 1.2255
Epoch [   14/   50] Time: 0.89min | d_loss: 0.6391 | g_loss: 1.4546
Epoch [   15/   50] Time: 0.00min | d_loss: 0.7789 | g_loss: 2.3187
Epoch [   15/   50] Time: 0.89min | d_loss: 0.6675 | g_loss: 1.4980
Epoch [   16/   50] Time: 0.00min | d_loss: 0.6288 | g_loss: 2.6071
Epoch [   16/   50] Time: 0.89min | d_loss: 0.7325 | g_loss: 2.1743
Epoch [   17/   50] Time: 0.00min | d_loss: 0.4222 | g_loss: 1.7040
Epoch [   17/   50] Time: 0.89min | d_loss: 0.6109 | g_loss: 2.6928
Epoch [   18/   50] Time: 0.00min | d_loss: 0.8011 | g_loss: 1.3176
Epoch [   18/   50] Time: 0.89min | d_loss: 0.8709 | g_loss: 2.0476
Epoch [   19/   50] Time: 0.00min | d_loss: 1.9739 | g_loss: 0.6493
Epoch [   19/   50] Time: 0.89min | d_loss: 0.6533 | g_loss: 2.3601
Epoch [   20/   50] Time: 0.00min | d_loss: 0.5412 | g_loss: 1.8728
Epoch [   20/   50] Time: 0.89min | d_loss: 0.6239 | g_loss: 2.4706
Epoch [   21/   50] Time: 0.00min | d_loss: 0.5969 | g_loss: 2.1323
Epoch [   21/   50] Time: 0.90min | d_loss: 0.5773 | g_loss: 2.1117
Epoch [   22/   50] Time: 0.00min | d_loss: 0.4986 | g_loss: 1.9018
Epoch [   22/   50] Time: 0.89min | d_loss: 0.4562 | g_loss: 2.4367
Epoch [   23/   50] Time: 0.00min | d_loss: 0.4909 | g_loss: 1.7718
Epoch [   23/   50] Time: 0.90min | d_loss: 0.4016 | g_loss: 2.3748
Epoch [   24/   50] Time: 0.00min | d_loss: 0.6217 | g_loss: 2.8365
Epoch [   24/   50] Time: 0.89min | d_loss: 1.5862 | g_loss: 1.0825
Epoch [   25/   50] Time: 0.00min | d_loss: 0.5001 | g_loss: 1.9359
Epoch [   25/   50] Time: 0.89min | d_loss: 0.3478 | g_loss: 2.4836
Epoch [   26/   50] Time: 0.00min | d_loss: 0.5571 | g_loss: 3.0008
Epoch [   26/   50] Time: 0.90min | d_loss: 0.3516 | g_loss: 2.2737
Epoch [   27/   50] Time: 0.00min | d_loss: 0.6012 | g_loss: 2.5301
Epoch [   27/   50] Time: 0.90min | d_loss: 0.4386 | g_loss: 1.2740
Epoch [   28/   50] Time: 0.00min | d_loss: 0.4990 | g_loss: 2.7584
Epoch [   28/   50] Time: 0.90min | d_loss: 0.2970 | g_loss: 3.0469
Epoch [   29/   50] Time: 0.00min | d_loss: 0.4103 | g_loss: 2.4090
Epoch [   29/   50] Time: 0.90min | d_loss: 0.2794 | g_loss: 2.9859
Epoch [   30/   50] Time: 0.00min | d_loss: 0.3412 | g_loss: 2.6635
Epoch [   30/   50] Time: 0.90min | d_loss: 0.3761 | g_loss: 2.9572
Epoch [   31/   50] Time: 0.00min | d_loss: 0.3321 | g_loss: 2.8182
Epoch [   31/   50] Time: 0.90min | d_loss: 0.3717 | g_loss: 2.2221
Epoch [   32/   50] Time: 0.00min | d_loss: 0.3235 | g_loss: 3.1171
Epoch [   32/   50] Time: 0.90min | d_loss: 0.6218 | g_loss: 2.8124
Epoch [   33/   50] Time: 0.00min | d_loss: 0.3382 | g_loss: 2.2436
Epoch [   33/   50] Time: 0.89min | d_loss: 0.2990 | g_loss: 2.3353
Epoch [   34/   50] Time: 0.00min | d_loss: 0.4286 | g_loss: 2.1040
Epoch [   34/   50] Time: 0.90min | d_loss: 0.2279 | g_loss: 3.3233
Epoch [   35/   50] Time: 0.00min | d_loss: 0.4569 | g_loss: 2.5604
Epoch [   35/   50] Time: 0.89min | d_loss: 0.2860 | g_loss: 3.9750
Epoch [   36/   50] Time: 0.00min | d_loss: 0.3356 | g_loss: 3.2076
Epoch [   36/   50] Time: 0.89min | d_loss: 0.2176 | g_loss: 2.8360
Epoch [   37/   50] Time: 0.00min | d_loss: 0.3764 | g_loss: 2.5194
Epoch [   37/   50] Time: 0.90min | d_loss: 0.2387 | g_loss: 3.8018
Epoch [   38/   50] Time: 0.00min | d_loss: 0.2708 | g_loss: 2.6302
Epoch [   38/   50] Time: 0.89min | d_loss: 0.2962 | g_loss: 2.6250
Epoch [   39/   50] Time: 0.00min | d_loss: 0.2434 | g_loss: 3.1644
Epoch [   39/   50] Time: 0.90min | d_loss: 0.8336 | g_loss: 1.5102
Epoch [   40/   50] Time: 0.00min | d_loss: 0.3295 | g_loss: 2.6875
Epoch [   40/   50] Time: 0.89min | d_loss: 0.2895 | g_loss: 4.1759
Epoch [   41/   50] Time: 0.00min | d_loss: 1.1815 | g_loss: 1.2811
Epoch [   41/   50] Time: 0.90min | d_loss: 0.9105 | g_loss: 1.9682
Epoch [   42/   50] Time: 0.00min | d_loss: 0.4744 | g_loss: 3.6863
Epoch [   42/   50] Time: 0.89min | d_loss: 0.2939 | g_loss: 3.6747
Epoch [   43/   50] Time: 0.00min | d_loss: 1.8112 | g_loss: 2.1459
Epoch [   43/   50] Time: 0.90min | d_loss: 0.2932 | g_loss: 3.8418
Epoch [   44/   50] Time: 0.00min | d_loss: 0.2255 | g_loss: 3.2089
Epoch [   44/   50] Time: 0.89min | d_loss: 0.4117 | g_loss: 3.9751
Epoch [   45/   50] Time: 0.00min | d_loss: 0.2330 | g_loss: 2.2559
Epoch [   45/   50] Time: 0.90min | d_loss: 0.2655 | g_loss: 2.5806
Epoch [   46/   50] Time: 0.00min | d_loss: 0.2098 | g_loss: 4.1591
Epoch [   46/   50] Time: 0.89min | d_loss: 0.1879 | g_loss: 4.1581
Epoch [   47/   50] Time: 0.00min | d_loss: 0.1590 | g_loss: 3.6443
Epoch [   47/   50] Time: 0.89min | d_loss: 0.3401 | g_loss: 3.6349
Epoch [   48/   50] Time: 0.00min | d_loss: 0.1512 | g_loss: 3.1256
Epoch [   48/   50] Time: 0.90min | d_loss: 0.1771 | g_loss: 2.8388
Epoch [   49/   50] Time: 0.00min | d_loss: 0.3822 | g_loss: 4.0616
Epoch [   49/   50] Time: 0.89min | d_loss: 0.1567 | g_loss: 3.2116
Epoch [   50/   50] Time: 0.00min | d_loss: 0.2945 | g_loss: 4.1574
Epoch [   50/   50] Time: 0.90min | d_loss: 0.1556 | g_loss: 3.8230

Training loss

Plot the training losses for the generator and discriminator, recorded after each epoch.

In [22]:
fig, ax = plt.subplots()
losses = np.array(losses)
plt.plot(losses.T[0], label='Discriminator', alpha=0.5)
plt.plot(losses.T[1], label='Generator', alpha=0.5)
plt.title("Training Losses")
plt.legend()
Out[22]:
<matplotlib.legend.Legend at 0x7f275abcda90>

Generator samples from training

View samples of images from the generator, and answer a question about the strengths and weaknesses of your trained models.

In [23]:
# Load samples from generator, taken while training
with open('train_samples.pkl', 'rb') as f:
    samples = pkl.load(f)
In [20]:
# helper function for viewing a list of passed in sample images
def view_samples(epoch, samples):
    fig, axes = plt.subplots(figsize=(16,4), nrows=2, ncols=8, sharey=True, sharex=True)
    for ax, img in zip(axes.flatten(), samples[epoch]):
        img = img.detach().cpu().numpy()
        img = np.transpose(img, (1, 2, 0))
        img = ((img + 1)*255 / (2)).astype(np.uint8)
        ax.xaxis.set_visible(False)
        ax.yaxis.set_visible(False)
        im = ax.imshow(img.reshape((32,32,3)))
    plt.show()
In [24]:
_ = view_samples(-1, samples)

Question: What do you notice about your generated samples and how might you improve this model?

When you answer this question, consider the following factors:

  • The dataset is biased; it is made of "celebrity" faces that are mostly white
  • Model size; larger models have the opportunity to learn more features in a data feature space
  • Optimization strategy; optimizers and number of epochs affect your final result

Answer: (Write your answer in this cell)

  • The generated sample images are blurry and not that much clear for gender also. If we increased the training image shape we can improve this.
  • We have only 200,000 training images so increasing images examples will definitely help improving the performance but more resources are needed.
  • I have used only 3 convolution layers with kernel size 4 and same number of transpose convolution layers adding some more layers and then training for some more epochs will sure give best result with bit more resources usage.
  • Human face is complex in feature extraction and our dataset have some unwanted features like hat, glasses microphone also which is affecting our performance. Also it seems that the model is becoming biased towards white skin.
  • About optimization, increasing the batch size and decreasing learning rate, beta1 and increasing epochs might tune the performance.

Submitting This Project

When submitting this project, make sure to run all the cells before saving the notebook. Save the notebook file as "dlnd_face_generation.ipynb" and save it as a HTML file under "File" -> "Download as". Include the "problem_unittests.py" files in your submission.